Update docs for dev (build 4264)

Author: Circle CI

dev/_downloads/07fcc19ba03226cd3d83d4e40ec44385/auto_examples_python.zip

@@ -69,10 +69,10 @@ To run the benchmark, just execute:
     <span style="color: #000080; text-decoration-color: #000080; font-weight: bold">    |--gd[lr=0.01]:</span> <span style="color: #808000; text-decoration-color: #808000; font-weight: bold">done (not enough run)</span>                                     
     <span style="color: #008000; text-decoration-color: #008000; font-weight: bold">Saving result in: </span>
     <span style="color: #008000; text-decoration-color: #008000; font-weight: bold">/home/circleci/project/examples/minimal_benchmark/outputs/benchopt_run_2026-03</span>
-    <span style="color: #008000; text-decoration-color: #008000; font-weight: bold">-01_19h47m53.parquet</span>
-    Processing 
+    <span style="color: #008000; text-decoration-color: #008000; font-weight: bold">-02_12h26m27.parquet</span>
+       Processing 
     /home/circleci/project/examples/minimal_benchmark/outputs/benchopt_run_2026-03
-    -01_19h47m53.parquet
+    -02_12h26m27.parquet
 
 
     </pre></div>
@@ -121,10 +121,10 @@ To get a more precise curve, you can increase ``n`` and ``r``:
     <span style="color: #000080; text-decoration-color: #000080; font-weight: bold">    |--gd[lr=0.01]:</span> <span style="color: #008000; text-decoration-color: #008000; font-weight: bold">done</span>                                                      
     <span style="color: #008000; text-decoration-color: #008000; font-weight: bold">Saving result in: </span>
     <span style="color: #008000; text-decoration-color: #008000; font-weight: bold">/home/circleci/project/examples/minimal_benchmark/outputs/benchopt_run_2026-03</span>
-    <span style="color: #008000; text-decoration-color: #008000; font-weight: bold">-01_19h48m08.parquet</span>
-    Processing 
+    <span style="color: #008000; text-decoration-color: #008000; font-weight: bold">-02_12h26m37.parquet</span>
+       Processing 
     /home/circleci/project/examples/minimal_benchmark/outputs/benchopt_run_2026-03
-    -01_19h48m08.parquet
+    -02_12h26m37.parquet
 
 
     </pre></div>
@@ -153,7 +153,7 @@ suboptimality instead of objective value.
 
 .. rst-class:: sphx-glr-timing
 
-   **Total running time of the script:** (0 minutes 19.234 seconds)
+   **Total running time of the script:** (0 minutes 14.948 seconds)
 
 
 .. _sphx_glr_download_auto_examples_run_minimal_benchmark.py:

dev/_downloads/586c9ec48a37fd397fb981a992a54250/run_minimal_benchmark.zip

@@ -6,7 +6,7 @@
 
 Computation times
 =================
-**00:19.234** total execution time for 3 files **from auto_examples**:
+**00:14.948** total execution time for 3 files **from auto_examples**:
 
 .. container::
 
@@ -33,7 +33,7 @@ Computation times
      - Time
      - Mem (MB)
    * - :ref:`sphx_glr_auto_examples_run_minimal_benchmark.py` (``run_minimal_benchmark.py``)
-     - 00:19.234
+     - 00:14.948
      - 0.0
    * - :ref:`sphx_glr_auto_examples_plot_run_benchmark.py` (``plot_run_benchmark.py``)
      - 00:00.000

dev/_downloads/6f1e7a639e0699d6164445b55e6c116d/auto_examples_jupyter.zip

@@ -23,21 +23,21 @@ The DataFrame contains the following columns:
   objective and solver, displayed in the HTML page.
 - ``file_objective|file_solver|file_dataset``: The filename of the objective,
   solver, and dataset used for this line.
-- ``p_obj_p``: the value of the objective's parameter ``p``.
-- ``p_solver_p``: the value of the solver's parameter ``p``.
-- ``p_dateset_p``: the value of the dataset's parameter ``p``.
+- ``p_obj_{p}``: the value of the objective's parameter ``p``.
+- ``p_solver_{p}``: the value of the solver's parameter ``p``.
+- ``p_dataset_{p}``: the value of the dataset's parameter ``p``.
 - ``time``: the time taken to run the solver until this point of the performance curve.
 - ``stop_val``: the number of iterations or the tolerance reached by the solver.
 - ``idx_rep``: If multiple repetitions are run for each solver with ``--n-rep``,
   this column contains the repetition number.
 - ``sampling_strategy``: The sampling strategy used to generate the performance
   curve of this solver. This allow to adapt the plot in the HTML depending on
   each solver.
-- ``objective_k``: The value associated to the key ``k`` in the
+- ``objective_{k}``: The value associated to the key ``k`` in the
   ``Objective.evaluate_result`` dictionary.
 
 The remaining columns are informations about the system used to run the
-benchmark, with keys ``{'env_OMP_NUM_THREADS', 'platform', 'platform-architecture', 'platform-release', 'platform-version', 'system-cpus', 'system-processor', 'system-ram (GB)', 'version-cuda', 'version-numpy', 'version-scipy', 'benchmark-git-tag'}``.
+benchmark, with keys ``{'env_OMP_NUM_THREADS', 'platform', 'platform-architecture', 'platform-release', 'platform-version', 'system-cpus', 'system-processor', 'system-ram (GB)', 'version-cuda', 'version-numpy', 'version-scipy', 'benchmark-git-tag', 'run_date'}``.
 
 
 .. _collect_results:
@@ -52,6 +52,30 @@ computed can be collected using the :option:`--collect` option with
 produce a parquet file with all the results that have been computed so far.
 
 
+.. _merge_results:
+
+Merge results from multiple runs
+--------------------------------
+
+Multiple runs of the same benchmark can be merged together using the
+``benchopt merge`` command. This allows to collect multiple runs, or to
+aggregate results from different users. By default, the merged results contain
+all lines from the different result files, with no additional processing.
+
+The resulting file is stored in the benchmark directory with the name
+``merged_results.parquet``. The name can be changed using the :option:`--output` option of the ``benchopt merge`` command.
+
+It is also possible to specify the ``--overwrite`` option to only keep one line
+per ``objective_name``, ``solver_name``, ``dataset_name``, and
+``idx_repetition``. In this case, only the line from the most recent run will
+be kept for each evaluation.
+
+Note that the run date is included in the result file in the ``run_date``
+columm. If it is not present in the original result file, benchopt uses
+the file creation date as a proxy for the run date. This allows to easily
+identify the most recent line when using the ``--overwrite`` option.
+
+
 Clean benchmark results
 -----------------------
 

dev/_downloads/a3747ad75224b0e1c78c72c3113052a8/plot_run_benchmark_python_R.zip

@@ -6,7 +6,7 @@
 
 Computation times
 =================
-**00:19.234** total execution time for 3 files **from all galleries**:
+**00:14.948** total execution time for 3 files **from all galleries**:
 
 .. container::
 
@@ -33,7 +33,7 @@ Computation times
      - Time
      - Mem (MB)
    * - :ref:`sphx_glr_auto_examples_run_minimal_benchmark.py` (``../examples/run_minimal_benchmark.py``)
-     - 00:19.234
+     - 00:14.948
      - 0.0
    * - :ref:`sphx_glr_auto_examples_plot_run_benchmark.py` (``../examples/plot_run_benchmark.py``)
      - 00:00.000

dev/_downloads/b6a0a50a7a6c020b6c5ce18f62b67968/plot_run_benchmark.zip

@@ -10,6 +10,13 @@ What's new
 Version 1.9 - In development
 ----------------------------
 
+CLI
+---
+
+- Add ``benchopt merge`` to merge results from multiple benchopt runs.
+  See :ref:`merge_results` for more details.
+  By `Thomas Moreau`_ (:gh:`892`)
+
 API
 ---
 
@@ -29,7 +36,7 @@ FIX
 - Fix failure in ``benchopt test`` when using ``get_seed`` in a class.
   By `Thomas Moreau`_ (:gh:`889`)
 
-- Fix conda env creation and ``--recreate`` option, ot include minimal reqs.
+- Fix conda env creation and ``--recreate`` option, to include minimal reqs.
   By `Thomas Moreau`_ (:gh:`891`)
 
 .. _changes_1_8_1: